Evaluating multiple classifiers for stock price direction prediction

The study focuses on support vector machines and artificial neural networks because they are the two methods predominantly used in the machine learning literature on financial time series forecasting.

“However, we focus on the two methods predominantly used in the ML literature on financial time series forecasting, SVM and ANN (Ballings, Van den Poel, Hespeels, & Gryp, 2015).”

Most previous studies on ensemble methods for stock-market prediction only ensemble a specific type of classifier or regressor, such as neural networks, decision trees, or support vector machines.

“Despite numerous works revealing the dominance of ensemble classifier over single classifier, most of these studies only ensemble a specific type of classifier or regressor for stock-market prediction, such as NN [18][19][20], DT [21,22] and SVM [12,23].”

Most previous studies on ensemble methods for stock-market predictions adopted the decrease variance approach (boosting or bagging) and experimented with data from one country.

“Also, most previous studies [12,19,21,22,[24][25][26][27][28][29][30], on ensemble methods for stock-market predictions adopted the decrease variance approach (boosting or bagging) and experimented with data from one country.”

The performance of ensemble methods depends on the domain of the dataset used for the study.

“Nevertheless, the study concluded that the performance of ensemble methods is dependent on the domain of the dataset used for the study.”

A comparison between bagging (BAG) and boosting (BOT) combination techniques by [12,21] revealed that the BAG technique outperformed the BOT technique.

“Furthermore, a comparison between bagging (BAG) and boosting (BOT) combination techniques by [12,21] revealed that the BAG technique outperformed the BOT technique.”

Most existing studies compared ensemble classifiers or ensemble regressors, but not both.

“most of the studies compared between ensemble classifiers [12, 18-20, 22, 23, 28] or regressors [21,30] machine learning algorithms, but not both.”

The variation in accuracy by classifier ensembles over different datasets shows that the accuracy of ensemble methods in stock-market prediction is dependent on the origin of the data being analysed.

“The variation in accuracy by classifier ensembles over different dataset shows that the accuracy of ensemble methods in stockmarket prediction is dependent on the origin of the data being analysed, which affirms literature [12,21].”